Results
By the demo day of our course, we finished all the functionalities planned in our project proposal, which included:
- Facial recognition and registration.
- Database for people information storage.
- Allocation implementation in the backend.
- Build backend on cloud (EC2).
- Elevators animation demo on the web.
- All interfaces between any adjacent function blocks.
Since the random allocations are not determined, it’s hard to say how much the performance is improved. However, as shown by the screenshots in Appendices (which was the same as the demo day), we can see the total time cost for random allocation was 14.4 seconds and average 9.7 seconds for each person. After our allocation strategy, the total time consumption was 9.6 seconds, and an average of 5.9 seconds for each person. Moreover, our facial recognition function handled almost any cases might happen when registering such as multiple face detection and no face detected. After registration, the facial recognition is capable to detect and our system can access such person’s information.
Future Works
The numbers and levels of elevators are fixed in our project for simplicity and universality. We can develop a system with changeable numbers of elevators and levels if needed for a specific building.
Besides, we used single threading in the animation demo. With multi-threading, we may demo a much more complicated scenario.
What’s more, although we built such Smart Elevators System and implemented the functions, we never got a chance to work with a physical elevator system in person. Thus, if one day we are building our system into reality, the technology to work on a physical elevator might be necessary to learn.
Besides, we used single threading in the animation demo. With multi-threading, we may demo a much more complicated scenario.
What’s more, although we built such Smart Elevators System and implemented the functions, we never got a chance to work with a physical elevator system in person. Thus, if one day we are building our system into reality, the technology to work on a physical elevator might be necessary to learn.
Conclusion
Overall, we have completed most of our projected features, and the algorithms have been proven to be effective. Multiple facial recognition allows us to identify multiple people in the lobby at the same time, and give a scheduling result accordingly. With an elevator set that supports the interfaces we have defined--which is not difficult to make, it is theoretically possible to move the simulation onto hardware.
Just like the labs we did earlier this semester, we have learned a lot of powerful tools and put them together to build a functional IoT system.
Just like the labs we did earlier this semester, we have learned a lot of powerful tools and put them together to build a functional IoT system.